
TarsosDSP, a Real-Time Audio Processing
Framework in Java
Joren Six
1,2
, Olmo Cornelis
2
, Marc Leman
1
1
University Ghent, IPEM, Sint-Pietersnieuwstraat 41, 9000, Gent, Belgium
2
University College Ghent, School of Arts, Jozef Kluyskensstraat 2, 9000 Gent, Belgium
ABSTRACT
This paper presents TarsosDSP, a framework for real-time audio analysis and processing. Most libraries and
frameworks offer either audio analysis and feature extraction or audio synthesis and processing. TarsosDSP
is one of a only a few frameworks that offers both analysis, processing and feature extraction in real-time, a
unique feature in the Java ecosystem. The framework contains practical audio processing algorithms, it can
be extended easily, and has no external dependencies. Each algorithm is implemented as simple as possible
thanks to a straightforward processing pipeline. TarsosDSP’s features include a resampling algorithm, onset
detectors, a number of pitch estimation algorithms, a time stretch algorithm, a pitch shifting algorithm, and
an algorithm to calculate the Constant-Q. The framework also allows simple audio synthesis, some audio
effects, and several filters. The Open Source framework is a valuable contribution to the MIR-Community
and ideal fit for interactive MIR-applications on Android.
1. INTRODUCTION
Frameworks or libraries
1
for audio processing can be di-
vided into two categories.The first category offers audio
analysis and feature extraction. The second category of-
fers audio synthesis capabilities. Both types may or may
not operate in real-time. Table 1 shows a partial overview
of notable audio frameworks. It shows that only a few
frameworks offer real-time feature extraction combined
with synthesis capabilities. To the best of the authors’
knowledge, TarsosDSP is unique in that regard within
the Java ecosystem. The combination of real-time fea-
ture extraction and synthesis can be of use for music ed-
ucation tools or music video games. Especially for de-
velopment on the Android platform there is a need for
such functionality.
TarsosDSP also fills a need for educational tools for
Music Information Retrieval. As identified by Gomez
in [14], there is a need for comprehensible, well-
documented MIR-frameworks which perform useful
tasks on every platforms, without the requirement of a
costly software package like Matlab. TarsosDSP serves
1
The distinction between library and framework is explained in [2].
In short, a framework is an abstract specification of an application
whereby analysis and design is reused, conversely when using a (class)
library code is reused but a library does not enforce a design.
this educational goal, it has already been used by several
master students as a starting point into music information
retrieval[5, 32, 28].
The framework tries to hit the sweet spot between be-
ing capable enough to get real tasks done, and compact
enough to serve as a demonstration for beginning MIR-
researchers on how audio processing works in practice.
TarsosDSP therefore targets both students and more ex-
perienced researchers who want to make use of the im-
plemented features.
After this introduction a section about the design deci-
sions made follows, then the main features of TarsosDSP
are highlighted. Chapter four is about the availability of
the framework. The paper ends with a conclusion and
future work.
2. DESIGN DECISIONS
To meet the goals stated in the introduction a couple of
design decisions were made.
2.1. Java based
TarsosDSP was written in Java to allow portability from
one platform to another. The automatic memory manage-
ment facilities are a great boon for a system implemented
in Java. These features allow a clean implementation
AES 53
RD
INTERNATIONAL CONFERENCE, London, UK, 2014 January 27–29
1